Uso de técnicas de processamento de linguagem natural e machine learning na criação de um modelo de recomendação de vagas de empregabilidade: um estudo de caso na plataforma mais acesso conecta

Abstract

The inclusion of People with Disabilities (PwD) in the labor market faces barriers that go beyond job availability, struggling with the efficient connection between profiles and opportuni ties. The Mais Acesso Conecta platform centralizes these job openings, but the search process relied on manual navigation and screening, being labor-intensive and inefficient for both can didates and recruiters. Given this limitation, it became necessary to develop a mechanism to automatically recommend similar opportunities aligned with the actual competencies of each profile. This work presents the development of a content-based recommendation system utilizing Natural Language Processing (NLP) and Machine Learning techniques. The adopted approach employs the Sentence-BERT contextual embeddings model in a Zero-Shot Learning strategy, eliminating the need for supervised training. The validation, performed through comparison with a statistical baseline (TF-IDF) and scenario analysis, demonstrated that the contextual vector approach is superior in semantic distinction, presenting a significant score distinction compared to noise vacancies, validating the efficacy of the proposed solution to intelligently expand employability